%autosave 0
Artificial Intelligence Nanodegree¶
Convolutional Neural Networks¶
Project: Write an Algorithm for a Dog Identification App¶
In this notebook, some template code has already been provided for you, and you will need to implement additional functionality to successfully complete this project. You will not need to modify the included code beyond what is requested. Sections that begin with '(IMPLEMENTATION)' in the header indicate that the following block of code will require additional functionality which you must provide. Instructions will be provided for each section, and the specifics of the implementation are marked in the code block with a 'TODO' statement. Please be sure to read the instructions carefully!
Note: Once you have completed all of the code implementations, you need to finalize your work by exporting the iPython Notebook as an HTML document. Before exporting the notebook to html, all of the code cells need to have been run so that reviewers can see the final implementation and output. You can then export the notebook by using the menu above and navigating to \n", "File -> Download as -> HTML (.html). Include the finished document along with this notebook as your submission.
In addition to implementing code, there will be questions that you must answer which relate to the project and your implementation. Each section where you will answer a question is preceded by a 'Question X' header. Carefully read each question and provide thorough answers in the following text boxes that begin with 'Answer:'. Your project submission will be evaluated based on your answers to each of the questions and the implementation you provide.
Note: Code and Markdown cells can be executed using the Shift + Enter keyboard shortcut. Markdown cells can be edited by double-clicking the cell to enter edit mode.
The rubric contains optional "Stand Out Suggestions" for enhancing the project beyond the minimum requirements. If you decide to pursue the "Stand Out Suggestions", you should include the code in this IPython notebook.
Why We're Here¶
In this notebook, you will make the first steps towards developing an algorithm that could be used as part of a mobile or web app. At the end of this project, your code will accept any user-supplied image as input. If a dog is detected in the image, it will provide an estimate of the dog's breed. If a human is detected, it will provide an estimate of the dog breed that is most resembling. The image below displays potential sample output of your finished project (... but we expect that each student's algorithm will behave differently!).
In this real-world setting, you will need to piece together a series of models to perform different tasks; for instance, the algorithm that detects humans in an image will be different from the CNN that infers dog breed. There are many points of possible failure, and no perfect algorithm exists. Your imperfect solution will nonetheless create a fun user experience!
The Road Ahead¶
We break the notebook into separate steps. Feel free to use the links below to navigate the notebook.
- Step 0: Import Datasets
- Step 1: Detect Humans
- Step 2: Detect Dogs
- Step 3: Create a CNN to Classify Dog Breeds (from Scratch)
- Step 4: Use a CNN to Classify Dog Breeds (using Transfer Learning)
- Step 5: Create a CNN to Classify Dog Breeds (using Transfer Learning)
- Step 6: Write your Algorithm
- Step 7: Test Your Algorithm
Step 0: Import Datasets¶
Import Dog Dataset¶
In the code cell below, we import a dataset of dog images. We populate a few variables through the use of the load_files function from the scikit-learn library:
train_files,valid_files,test_files- numpy arrays containing file paths to imagestrain_targets,valid_targets,test_targets- numpy arrays containing onehot-encoded classification labelsdog_names- list of string-valued dog breed names for translating labels
from sklearn.datasets import load_files
from keras.utils import np_utils
import numpy as np
from glob import glob
# define function to load train, test, and validation datasets
def load_dataset(path):
data = load_files(path)
dog_files = np.array(data['filenames'])
dog_targets = np_utils.to_categorical(np.array(data['target']), 133)
return dog_files, dog_targets
# load train, test, and validation datasets
train_files, train_targets = load_dataset('dogImages/train')
valid_files, valid_targets = load_dataset('dogImages/valid')
test_files, test_targets = load_dataset('dogImages/test')
# load list of dog names
dog_names = [item[20:-1] for item in sorted(glob("dogImages/train/*/"))]
# print statistics about the dataset
print('There are %d total dog categories.' % len(dog_names))
print('There are %s total dog images.\n' % len(np.hstack([train_files, valid_files, test_files])))
print('There are %d training dog images.' % len(train_files))
print('There are %d validation dog images.' % len(valid_files))
print('There are %d test dog images.'% len(test_files))
print(train_files[2], '\n', train_targets[2])
Import Human Dataset¶
In the code cell below, we import a dataset of human images, where the file paths are stored in the numpy array human_files.
import random
random.seed(853092)
# load filenames in shuffled human dataset
human_files = np.array(glob("lfw/*/*"))
random.shuffle(human_files)
# print statistics about the dataset
print('There are %d total human images.' % len(human_files))
Step 1: Detect Humans¶
We use OpenCV's implementation of Haar feature-based cascade classifiers to detect human faces in images. OpenCV provides many pre-trained face detectors, stored as XML files on github. We have downloaded one of these detectors and stored it in the haarcascades directory.
In the next code cell, we demonstrate how to use this detector to find human faces in a sample image.
import cv2
import cv2
import matplotlib.pyplot as plt
%matplotlib inline
# extract pre-trained face detector
face_cascade = cv2.CascadeClassifier('haarcascades/haarcascade_frontalface_alt.xml')
# load color (BGR) image
img = cv2.imread(human_files[19])
# convert BGR image to grayscale
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
# find faces in image
faces = face_cascade.detectMultiScale(gray)
# print number of faces detected in the image
print('Number of faces detected:', len(faces))
# get bounding box for each detected face
for (x,y,w,h) in faces:
# add bounding box to color image
cv2.rectangle(img,(x,y),(x+w,y+h),(255,0,0),2)
# convert BGR image to RGB for plotting
cv_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# display the image, along with bounding box
plt.imshow(cv_rgb)
plt.show()
#print ('imagedata\n',gray.shape)
Before using any of the face detectors, it is standard procedure to convert the images to grayscale. The detectMultiScale function executes the classifier stored in face_cascade and takes the grayscale image as a parameter.
In the above code, faces is a numpy array of detected faces, where each row corresponds to a detected face. Each detected face is a 1D array with four entries that specifies the bounding box of the detected face. The first two entries in the array (extracted in the above code as x and y) specify the horizontal and vertical positions of the top left corner of the bounding box. The last two entries in the array (extracted here as w and h) specify the width and height of the box.
Write a Human Face Detector¶
We can use this procedure to write a function that returns True if a human face is detected in an image and False otherwise. This function, aptly named face_detector, takes a string-valued file path to an image as input and appears in the code block below.
# returns "True" if face is detected in image stored at img_path
def face_detector(img_path):
img = cv2.imread(img_path)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
faces = face_cascade.detectMultiScale(gray)
if len(faces) == 0 and img_path.startswith('lfw'):
print(img_path)
return len(faces) > 0
(IMPLEMENTATION) Assess the Human Face Detector¶
Question 1: Use the code cell below to test the performance of the face_detector function.
- What percentage of the first 100 images in
human_fileshave a detected human face? - What percentage of the first 100 images in
dog_fileshave a detected human face?
Ideally, we would like 100% of human images with a detected face and 0% of dog images with a detected face. You will see that our algorithm falls short of this goal, but still gives acceptable performance. We extract the file paths for the first 100 images from each of the datasets and store them in the numpy arrays human_files_short and dog_files_short.
Answer:
- What percentage of the first 100 images in
human_fileshave a detected human face?The human face detector works accurately 99% of the time, misses 1 face (idx=6, as seen above). - What percentage of the first 100 images in
dog_fileshave a detected human face?11% of the dog faces are incorrectly identified as human faces.
human_files_short = human_files[:100]
dog_files_short = train_files[:100]
# Do NOT modify the code above this line.
#print (face_detector(dog_files_short[10]))
## TODO: Test the performance of the face_detector algorithm
## on the images in human_files_short and dog_files_short.
h_det = list(map(face_detector, human_files_short))
d_det = list(map(face_detector, dog_files_short))
print ('{}\n{}'.format(sum(h_det),sum(d_det)))
Question 2: This algorithmic choice necessitates that we communicate to the user that we accept human images only when they provide a clear view of a face (otherwise, we risk having unneccessarily frustrated users!). In your opinion, is this a reasonable expectation to pose on the user? If not, can you think of a way to detect humans in images that does not necessitate an image with a clearly presented face?
Answer:
In the real world, this is not a reasonable request. A human can recognize a face even without a clear view of the face, e.g. sideways, tilted, or even upside-down.
For our checkimg, let's try out an mlp first, and then the CNN (using models as-is from the aind-cnn project)
We suggest the face detector from OpenCV as a potential way to detect human images in your algorithm, but you are free to explore other approaches, especially approaches that make use of deep learning :). Please use the code cell below to design and test your own face detection algorithm. If you decide to pursue this optional task, report performance on each of the datasets.
Reference: https://hjweide.github.io/efficient-image-loading
## (Optional) TODO: Report the performance of another
## face detection algorithm on the LFW dataset
### Feel free to use as many code cells as needed.
# First the MLP
# We need to load the human faces into x_train, y_train, x_test, y_test
#cv2.CV_LOAD_IMAGE_GRAYSCALE has an integer value of 0. use that in imread
def getX(image_dir):
data = np.empty((len(image_dir), 62500), dtype=np.uint8)
labels = np.empty((len(image_dir), 2), dtype=np.uint8)
for i, fpath in enumerate(image_dir):
#cv2.CV_LOAD_IMAGE_GRAYSCALE has an integer value of 0. use that in imread
img = cv2.imread(fpath, 0)
if img.shape is not (250, 250):
img = cv2.resize(img, (250,250))
img = img.flatten()
data[i, ...] = img
labels[i] = [1,0] if 'lfw' in fpath else [0, 1]
return data,labels
def load_human_and_dog_faces(tr=100, vl=25, ts=50):
tr_data, tr_labels = getX(np.concatenate([human_files[:tr], train_files[:tr]]))
vl_data, vl_labels = getX(np.concatenate([human_files[tr:tr+vl], train_files[tr:tr+vl]]))
ts_data, ts_labels = getX(np.concatenate([human_files[tr+vl:tr+vl+ts], train_files[tr+vl:tr+vl+ts]]))
return tr_data, tr_labels, vl_data, vl_labels, ts_data, ts_labels
x_train, y_train, x_valid, y_valid, x_test, y_test = load_human_and_dog_faces(tr=500, vl=50, ts=50)
print(x_train.shape, y_train.shape, x_valid.shape, y_valid.shape, x_test.shape, y_test.shape)
num_classes = 2
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten
# define the model
model = Sequential()
model.add(Dense(6250, input_shape=(62500,), activation='relu'))
model.add(Dense(512, activation='relu'))
model.add(Dropout(0.8))
model.add(Dense(512, activation='relu'))
model.add(Dropout(0.1))
model.add(Dense(num_classes, activation='softmax'))
# compile the model
model.compile(loss='categorical_crossentropy', optimizer='rmsprop',
metrics=['accuracy'])
run_fit = False #set to True when ready to run fit()
#model.summary()
from keras.callbacks import ModelCheckpoint
if run_fit:
# train the model, and save the best
checkpointer = ModelCheckpoint(filepath='saved_models/MLP.human.weights.best.hdf5', verbose=1,
save_best_only=True)
hist = model.fit(x_train, y_train, batch_size=32, epochs=2,
validation_data=(x_valid, y_valid), callbacks=[checkpointer],
verbose=2, shuffle=True)
# load the weights that yielded the best validation accuracy
model.load_weights('saved_models/MLP.human.weights.best.hdf5')
# evaluate and print test accuracy
score = model.evaluate(x_test, y_test, verbose=0)
print('\n', 'Test accuracy:', score[1])
(1000, 62500) (1000, 2) (100, 62500) (100, 2) (100, 62500) (100, 2) _________________________________________________________________ Layer (type) Output Shape Param # ================================================================= dense_1 (Dense) (None, 6250) 390631250 _________________________________________________________________ dense_2 (Dense) (None, 512) 3200512 _________________________________________________________________ dropout_1 (Dropout) (None, 512) 0 _________________________________________________________________ dense_3 (Dense) (None, 512) 262656 _________________________________________________________________ dropout_2 (Dropout) (None, 512) 0 _________________________________________________________________ dense_4 (Dense) (None, 2) 1026 ================================================================= Total params: 394,095,444.0 Trainable params: 394,095,444.0 Non-trainable params: 0.0 _________________________________________________________________ Train on 1000 samples, validate on 100 samples Epoch 1/2 Epoch 00000: val_loss improved from inf to 8.05905, saving model to MLP.human.weights.best.hdf5 650s - loss: 8.1076 - acc: 0.4970 - val_loss: 8.0590 - val_acc: 0.5000 Epoch 2/2 Epoch 00001: val_loss did not improve 186s - loss: 8.0752 - acc: 0.4990 - val_loss: 8.0590 - val_acc: 0.5000 Test accuracy: 0.5
# Then the Conv2D
#Load the images in as color 3D because Conv2D can handle multi-dim inputs while perceptron needs a vector
def getX(image_dir):
data = np.empty((len(image_dir), 250, 250, 3), dtype=np.uint8)
labels = np.empty((len(image_dir), 2), dtype=np.uint8)
for i, fpath in enumerate(image_dir):
img = cv2.imread(fpath)
if img.shape[0] is not 250 or img.shape[1] is not 250:
img = cv2.resize(img, (250,250))
#img = img.transpose(2, 0, 1)
data[i, ...] = img
labels[i] = [1,0] if 'lfw' in fpath else [0, 1]
return data,labels
def load_human_and_dog_faces(tr=100, vl=25, ts=50):
tr_data, tr_labels = getX(np.concatenate([human_files[:tr], train_files[:tr]]))
vl_data, vl_labels = getX(np.concatenate([human_files[tr:tr+vl], train_files[tr:tr+vl]]))
ts_data, ts_labels = getX(np.concatenate([human_files[tr+vl:tr+vl+ts], train_files[tr+vl:tr+vl+ts]]))
return tr_data, tr_labels, vl_data, vl_labels, ts_data, ts_labels
x_train, y_train, x_valid, y_valid, x_test, y_test = load_human_and_dog_faces(tr=500, vl=50, ts=50)
#testpath='lfw/Harrison_Ford/Harrison_Ford_0003.jpg'
#testpath = 'dogImages/train/001.Affenpinscher/Affenpinscher_00001.jpg'
#print(testpath)
#img = cv2.imread(testpath)
#print(img.shape)
#print(img.shape[0] is not 250 and img.shape[1] is not 250)
#img = cv2.resize(img,(250,250))
#print(img.shape)
#img = img.transpose(2, 0, 1)
#print(img.shape)
#print(cv2.resize(img,(250,250)).shape)
#print(img)
print(x_train.shape, y_train.shape)
print(x_train.shape, y_train.shape)
from keras.models import Sequential
from keras.layers import Conv2D, MaxPooling2D, Flatten, Dense, Dropout
model = Sequential()
model.add(Conv2D(filters=16, kernel_size=2, padding='same', activation='relu',
input_shape=(250, 250, 3)))
model.add(MaxPooling2D(pool_size=2))
model.add(Conv2D(filters=32, kernel_size=2, padding='same', activation='relu'))
model.add(MaxPooling2D(pool_size=2))
model.add(Conv2D(filters=64, kernel_size=2, padding='same', activation='relu'))
model.add(MaxPooling2D(pool_size=2))
model.add(Dropout(0.3))
model.add(Flatten())
model.add(Dense(500, activation='relu'))
model.add(Dropout(0.4))
model.add(Dense(2, activation='softmax'))
model.summary()
# compile the model
model.compile(loss='categorical_crossentropy', optimizer='rmsprop',
metrics=['accuracy'])
from keras.preprocessing.image import ImageDataGenerator
# create and configure augmented image generator
datagen_train = ImageDataGenerator(
width_shift_range=0.1, # randomly shift images horizontally (10% of total width)
height_shift_range=0.1, # randomly shift images vertically (10% of total height)
horizontal_flip=True) # randomly flip images horizontally
# create and configure augmented image generator
datagen_valid = ImageDataGenerator(
width_shift_range=0.1, # randomly shift images horizontally (10% of total width)
height_shift_range=0.1, # randomly shift images vertically (10% of total height)
horizontal_flip=True) # randomly flip images horizontally
# fit augmented image generator on data
datagen_train.fit(x_train)
datagen_valid.fit(x_valid)
# DO NOT re-run this cell. Takes a very long time. Output is described in the cell below.
from keras.callbacks import ModelCheckpoint
batch_size = 32
epochs = 50
# train the model
checkpointer = ModelCheckpoint(filepath='saved_models/aug_model.human.weights.best.hdf5', verbose=1,
save_best_only=True)
model.fit_generator(datagen_train.flow(x_train, y_train, batch_size=batch_size),
steps_per_epoch=x_train.shape[0] // batch_size,
epochs=epochs, verbose=2, callbacks=[checkpointer],
validation_data=datagen_valid.flow(x_valid, y_valid, batch_size=batch_size),
validation_steps=x_valid.shape[0] // batch_size)
Epoch 1/100 Epoch 00000: val_loss improved from inf to 7.72325, saving model to aug_model.human.weights.best.hdf5 53s - loss: 8.1648 - acc: 0.4909 - val_loss: 7.7233 - val_acc: 0.5208 Epoch 2/100 Epoch 00001: val_loss did not improve 47s - loss: 8.0281 - acc: 0.5019 - val_loss: 8.5331 - val_acc: 0.4706 Epoch 3/100 Epoch 00002: val_loss improved from 7.72325 to 7.58499, saving model to aug_model.human.weights.best.hdf5 48s - loss: 7.9452 - acc: 0.5071 - val_loss: 7.5850 - val_acc: 0.5294 Epoch 4/100 Epoch 00003: val_loss did not improve 47s - loss: 8.0590 - acc: 0.5000 - val_loss: 7.5850 - val_acc: 0.5294 Epoch 5/100 Epoch 00004: val_loss did not improve 47s - loss: 8.1210 - acc: 0.4962 - val_loss: 8.0590 - val_acc: 0.5000 Epoch 6/100 Epoch 00005: val_loss did not improve 46s - loss: 8.0769 - acc: 0.4989 - val_loss: 8.0590 - val_acc: 0.5000 Epoch 7/100 Epoch 00006: val_loss did not improve 47s - loss: 7.9940 - acc: 0.5040 - val_loss: 8.0590 - val_acc: 0.5000 Epoch 8/100 Epoch 00007: val_loss improved from 7.58499 to 6.87389, saving model to aug_model.human.weights.best.hdf5 50s - loss: 8.1551 - acc: 0.4940 - val_loss: 6.8739 - val_acc: 0.5735 Epoch 9/100 Epoch 00008: val_loss did not improve 46s - loss: 8.1094 - acc: 0.4969 - val_loss: 8.2961 - val_acc: 0.4853 Epoch 10/100 Epoch 00009: val_loss did not improve 47s - loss: 7.9305 - acc: 0.5080 - val_loss: 7.8912 - val_acc: 0.5104 Epoch 11/100 Epoch 00010: val_loss did not improve 47s - loss: 8.2557 - acc: 0.4878 - val_loss: 9.7183 - val_acc: 0.3971 Epoch 12/100 Epoch 00011: val_loss did not improve 47s - loss: 7.9274 - acc: 0.5082 - val_loss: 8.0590 - val_acc: 0.5000 Epoch 13/100 Epoch 00012: val_loss did not improve 47s - loss: 7.9599 - acc: 0.5061 - val_loss: 8.2961 - val_acc: 0.4853 Epoch 14/100 Epoch 00013: val_loss did not improve 47s - loss: 8.3030 - acc: 0.4849 - val_loss: 7.5850 - val_acc: 0.5294 Epoch 15/100 Epoch 00014: val_loss did not improve 47s - loss: 7.9777 - acc: 0.5050 - val_loss: 8.0590 - val_acc: 0.5000 Epoch 16/100 Epoch 00015: val_loss did not improve 47s - loss: 8.0281 - acc: 0.5019 - val_loss: 8.5331 - val_acc: 0.4706 Epoch 17/100 Epoch 00016: val_loss did not improve 47s - loss: 7.8639 - acc: 0.5121 - val_loss: 6.8739 - val_acc: 0.5735 Epoch 18/100 Epoch 00017: val_loss did not improve 47s - loss: 8.3161 - acc: 0.4841 - val_loss: 7.1109 - val_acc: 0.5588 Epoch 19/100 Epoch 00018: val_loss did not improve 47s - loss: 8.0281 - acc: 0.5019 - val_loss: 7.3480 - val_acc: 0.5441 Epoch 20/100 Epoch 00019: val_loss did not improve 47s - loss: 7.9615 - acc: 0.5061 - val_loss: 8.2961 - val_acc: 0.4853 Epoch 21/100 Epoch 00020: val_loss did not improve 47s - loss: 7.8933 - acc: 0.5103 - val_loss: 6.8739 - val_acc: 0.5735 Epoch 22/100 Epoch 00021: val_loss did not improve 48s - loss: 8.5290 - acc: 0.4708 - val_loss: 7.5850 - val_acc: 0.5294 Epoch 23/100 Epoch 00022: val_loss did not improve 52s - loss: 7.8833 - acc: 0.5109 - val_loss: 7.8220 - val_acc: 0.5147 Epoch 24/100 Epoch 00023: val_loss did not improve 48s - loss: 7.9483 - acc: 0.5069 - val_loss: 9.2442 - val_acc: 0.4265 Epoch 25/100 Epoch 00024: val_loss did not improve 49s - loss: 8.0590 - acc: 0.5000 - val_loss: 7.8220 - val_acc: 0.5147 Epoch 26/100 Epoch 00025: val_loss did not improve 50s - loss: 8.0087 - acc: 0.5031 - val_loss: 7.3480 - val_acc: 0.5441 Epoch 27/100 Epoch 00026: val_loss did not improve 52s - loss: 7.9452 - acc: 0.5071 - val_loss: 8.2961 - val_acc: 0.4853 Epoch 28/100 Epoch 00027: val_loss did not improve 50s - loss: 8.2410 - acc: 0.4887 - val_loss: 8.3948 - val_acc: 0.4792 Epoch 29/100 Epoch 00028: val_loss did not improve 50s - loss: 8.0869 - acc: 0.4983 - val_loss: 8.0590 - val_acc: 0.5000 Epoch 30/100 Epoch 00029: val_loss did not improve 50s - loss: 7.9924 - acc: 0.5041 - val_loss: 9.0072 - val_acc: 0.4412 Epoch 31/100 Epoch 00030: val_loss did not improve 50s - loss: 8.0134 - acc: 0.5028 - val_loss: 7.8220 - val_acc: 0.5147 Epoch 32/100 Epoch 00031: val_loss did not improve 50s - loss: 8.1860 - acc: 0.4921 - val_loss: 7.3480 - val_acc: 0.5441 Epoch 33/100 Epoch 00032: val_loss did not improve 51s - loss: 8.0266 - acc: 0.5020 - val_loss: 8.7701 - val_acc: 0.4559 Epoch 34/100 Epoch 00033: val_loss did not improve 50s - loss: 8.2495 - acc: 0.4882 - val_loss: 8.0590 - val_acc: 0.5000 Epoch 35/100 Epoch 00034: val_loss did not improve 50s - loss: 7.9956 - acc: 0.5039 - val_loss: 9.2442 - val_acc: 0.4265 Epoch 36/100 Epoch 00035: val_loss did not improve 50s - loss: 8.1713 - acc: 0.4930 - val_loss: 8.7701 - val_acc: 0.4559 Epoch 37/100 Epoch 00036: val_loss did not improve 52s - loss: 8.1047 - acc: 0.4972 - val_loss: 7.5850 - val_acc: 0.5294 Epoch 38/100 Epoch 00037: val_loss did not improve 51s - loss: 8.0606 - acc: 0.4999 - val_loss: 7.5850 - val_acc: 0.5294 Epoch 39/100 Epoch 00038: val_loss did not improve 50s - loss: 8.1388 - acc: 0.4951 - val_loss: 8.7701 - val_acc: 0.4559 Epoch 40/100 Epoch 00039: val_loss did not improve 50s - loss: 7.8020 - acc: 0.5159 - val_loss: 7.8220 - val_acc: 0.5147 Epoch 41/100 Epoch 00040: val_loss did not improve 51s - loss: 8.2395 - acc: 0.4888 - val_loss: 9.0072 - val_acc: 0.4412 Epoch 42/100 Epoch 00041: val_loss did not improve 51s - loss: 7.9793 - acc: 0.5049 - val_loss: 8.2961 - val_acc: 0.4853 Epoch 43/100 Epoch 00042: val_loss did not improve 51s - loss: 8.0265 - acc: 0.5020 - val_loss: 8.2961 - val_acc: 0.4853 Epoch 44/100 Epoch 00043: val_loss improved from 6.87389 to 6.39983, saving model to aug_model.human.weights.best.hdf5 52s - loss: 7.8345 - acc: 0.5139 - val_loss: 6.3998 - val_acc: 0.6029 Epoch 45/100 Epoch 00044: val_loss did not improve 51s - loss: 8.0916 - acc: 0.4980 - val_loss: 7.3480 - val_acc: 0.5441 Epoch 46/100 Epoch 00045: val_loss did not improve 51s - loss: 8.0118 - acc: 0.5029 - val_loss: 7.8220 - val_acc: 0.5147 Epoch 47/100 Epoch 00046: val_loss did not improve 51s - loss: 8.1566 - acc: 0.4939 - val_loss: 8.7701 - val_acc: 0.4559 Epoch 48/100 Epoch 00047: val_loss did not improve 51s - loss: 8.2511 - acc: 0.4881 - val_loss: 8.5331 - val_acc: 0.4706 Epoch 49/100 Epoch 00048: val_loss did not improve 51s - loss: 7.8020 - acc: 0.5159 - val_loss: 8.5331 - val_acc: 0.4706 Epoch 50/100 Epoch 00049: val_loss did not improve 51s - loss: 8.0622 - acc: 0.4998 - val_loss: 7.5850 - val_acc: 0.5294 Epoch 51/100 Epoch 00050: val_loss did not improve 51s - loss: 7.9321 - acc: 0.5079 - val_loss: 7.5850 - val_acc: 0.5294 Epoch 52/100 Epoch 00051: val_loss did not improve 51s - loss: 8.0265 - acc: 0.5020 - val_loss: 8.0590 - val_acc: 0.5000 Epoch 53/100 Epoch 00052: val_loss did not improve 51s - loss: 8.1063 - acc: 0.4971 - val_loss: 8.2961 - val_acc: 0.4853 Epoch 54/100 Epoch 00053: val_loss did not improve 51s - loss: 7.7501 - acc: 0.5192 - val_loss: 8.2961 - val_acc: 0.4853 Epoch 55/100 Epoch 00054: val_loss did not improve 51s - loss: 8.2689 - acc: 0.4870 - val_loss: 8.7701 - val_acc: 0.4559 Epoch 56/100 Epoch 00055: val_loss did not improve 52s - loss: 8.3014 - acc: 0.4850 - val_loss: 8.7701 - val_acc: 0.4559 Epoch 57/100 Epoch 00056: val_loss did not improve 58s - loss: 7.9599 - acc: 0.5061 - val_loss: 8.5331 - val_acc: 0.4706 Epoch 58/100 Epoch 00057: val_loss did not improve 54s - loss: 8.0281 - acc: 0.5019 - val_loss: 7.8220 - val_acc: 0.5147 Epoch 59/100 Epoch 00058: val_loss did not improve 54s - loss: 7.8980 - acc: 0.5100 - val_loss: 7.5850 - val_acc: 0.5294 Epoch 60/100 Epoch 00059: val_loss did not improve 54s - loss: 8.4299 - acc: 0.4770 - val_loss: 7.3480 - val_acc: 0.5441 Epoch 61/100 Epoch 00060: val_loss did not improve 57s - loss: 7.9599 - acc: 0.5061 - val_loss: 8.2961 - val_acc: 0.4853 Epoch 62/100 Epoch 00061: val_loss did not improve 55s - loss: 8.0753 - acc: 0.4990 - val_loss: 6.8739 - val_acc: 0.5735 Epoch 63/100 Epoch 00062: val_loss did not improve 51s - loss: 8.0265 - acc: 0.5020 - val_loss: 7.8220 - val_acc: 0.5147 Epoch 64/100 Epoch 00063: val_loss did not improve 53s - loss: 8.2348 - acc: 0.4891 - val_loss: 8.0590 - val_acc: 0.5000 Epoch 65/100 Epoch 00064: val_loss did not improve 54s - loss: 8.0753 - acc: 0.4990 - val_loss: 8.0590 - val_acc: 0.5000 Epoch 66/100 Epoch 00065: val_loss did not improve 53s - loss: 7.9793 - acc: 0.5049 - val_loss: 7.8220 - val_acc: 0.5147 Epoch 67/100 Epoch 00066: val_loss did not improve 55s - loss: 8.2201 - acc: 0.4900 - val_loss: 7.5850 - val_acc: 0.5294 Epoch 68/100 Epoch 00067: val_loss did not improve 51s - loss: 8.0590 - acc: 0.5000 - val_loss: 8.0590 - val_acc: 0.5000 Epoch 69/100 Epoch 00068: val_loss did not improve 54s - loss: 8.0412 - acc: 0.5011 - val_loss: 8.7701 - val_acc: 0.4559 Epoch 70/100 Epoch 00069: val_loss did not improve 54s - loss: 7.8996 - acc: 0.5099 - val_loss: 7.8220 - val_acc: 0.5147 Epoch 71/100 Epoch 00070: val_loss did not improve 56s - loss: 8.0606 - acc: 0.4999 - val_loss: 7.1109 - val_acc: 0.5588 Epoch 72/100 Epoch 00071: val_loss did not improve 57s - loss: 8.0737 - acc: 0.4991 - val_loss: 7.8220 - val_acc: 0.5147 Epoch 73/100 Epoch 00072: val_loss did not improve 53s - loss: 8.0769 - acc: 0.4989 - val_loss: 8.0590 - val_acc: 0.5000 Epoch 74/100 Epoch 00073: val_loss did not improve 57s - loss: 8.1078 - acc: 0.4970 - val_loss: 6.8739 - val_acc: 0.5735 Epoch 75/100 Epoch 00074: val_loss did not improve 54s - loss: 8.0753 - acc: 0.4990 - val_loss: 8.7701 - val_acc: 0.4559 Epoch 76/100 Epoch 00075: val_loss did not improve 56s - loss: 8.0296 - acc: 0.5018 - val_loss: 7.5850 - val_acc: 0.5294 Epoch 77/100 Epoch 00076: val_loss did not improve 56s - loss: 8.0250 - acc: 0.5021 - val_loss: 8.2961 - val_acc: 0.4853 Epoch 78/100 Epoch 00077: val_loss did not improve 58s - loss: 7.9158 - acc: 0.5089 - val_loss: 9.0072 - val_acc: 0.4412 Epoch 79/100 Epoch 00078: val_loss did not improve 57s - loss: 8.2689 - acc: 0.4870 - val_loss: 8.5331 - val_acc: 0.4706 Epoch 80/100 Epoch 00079: val_loss did not improve 55s - loss: 7.7013 - acc: 0.5222 - val_loss: 7.8220 - val_acc: 0.5147 Epoch 81/100 Epoch 00080: val_loss did not improve 56s - loss: 8.3649 - acc: 0.4810 - val_loss: 8.2961 - val_acc: 0.4853 Epoch 82/100 Epoch 00081: val_loss did not improve 57s - loss: 7.8980 - acc: 0.5100 - val_loss: 6.8739 - val_acc: 0.5735 Epoch 83/100 Epoch 00082: val_loss did not improve 57s - loss: 7.6866 - acc: 0.5231 - val_loss: 7.3480 - val_acc: 0.5441 Epoch 84/100 Epoch 00083: val_loss did not improve 56s - loss: 8.3502 - acc: 0.4819 - val_loss: 9.0072 - val_acc: 0.4412 Epoch 85/100 Epoch 00084: val_loss did not improve 58s - loss: 8.1450 - acc: 0.4947 - val_loss: 8.2961 - val_acc: 0.4853 Epoch 86/100 Epoch 00085: val_loss did not improve 55s - loss: 7.7354 - acc: 0.5201 - val_loss: 9.0072 - val_acc: 0.4412 Epoch 87/100 Epoch 00086: val_loss did not improve 52s - loss: 8.2217 - acc: 0.4899 - val_loss: 8.7701 - val_acc: 0.4559 Epoch 88/100 Epoch 00087: val_loss did not improve 55s - loss: 8.2867 - acc: 0.4859 - val_loss: 9.0072 - val_acc: 0.4412 Epoch 89/100 Epoch 00088: val_loss did not improve 56s - loss: 7.9909 - acc: 0.5042 - val_loss: 8.0590 - val_acc: 0.5000 Epoch 90/100 Epoch 00089: val_loss did not improve 54s - loss: 8.0916 - acc: 0.4980 - val_loss: 8.0590 - val_acc: 0.5000 Epoch 91/100 Epoch 00090: val_loss did not improve 52s - loss: 7.8639 - acc: 0.5121 - val_loss: 8.0590 - val_acc: 0.5000 Epoch 92/100 Epoch 00091: val_loss did not improve 54s - loss: 8.1241 - acc: 0.4960 - val_loss: 8.2961 - val_acc: 0.4853 Epoch 93/100 Epoch 00092: val_loss did not improve 54s - loss: 8.2836 - acc: 0.4861 - val_loss: 8.5331 - val_acc: 0.4706 Epoch 94/100 Epoch 00093: val_loss did not improve 54s - loss: 7.9762 - acc: 0.5051 - val_loss: 8.0590 - val_acc: 0.5000 Epoch 95/100 Epoch 00094: val_loss did not improve 57s - loss: 8.1404 - acc: 0.4950 - val_loss: 8.5331 - val_acc: 0.4706 Epoch 96/100 Epoch 00095: val_loss did not improve 53s - loss: 8.0737 - acc: 0.4991 - val_loss: 7.3480 - val_acc: 0.5441 Epoch 97/100 Epoch 00096: val_loss did not improve 54s - loss: 8.0428 - acc: 0.5010 - val_loss: 8.5331 - val_acc: 0.4706 Epoch 98/100 Epoch 00097: val_loss did not improve 54s - loss: 8.0412 - acc: 0.5011 - val_loss: 8.2961 - val_acc: 0.4853 Epoch 99/100 Epoch 00098: val_loss did not improve 53s - loss: 8.1713 - acc: 0.4930 - val_loss: 8.5331 - val_acc: 0.4706 Epoch 100/100 Epoch 00099: val_loss did not improve 57s - loss: 8.1257 - acc: 0.4959 - val_loss: 8.2961 - val_acc: 0.4853 Out[29]:
# load the weights that yielded the best validation accuracy
model.load_weights('saved_models/aug_model.human.weights.best.hdf5')
# evaluate and print test accuracy
score = model.evaluate(x_test, y_test, verbose=0)
print('\n', 'Test accuracy:', score[1])
Results:
The MLP seems to take up a really large number of parameters, but quickly settles on to a val_loss value (8.0590) that does not improve over subsequent epochs beyond the 2nd epoch. This I tried with various combinations of 3-deep networks. None of the trials gives loss below 8.0590, or test accuracy better than 0.5, which is really the same as mindlessly predicting all the images to be of one class or the other. The accuracy does not change either with different shuffling of images, or by increasing the training data from 200 to 500 images.
Results of the Conv2D, even thought the loss is much lesser at 6.39983, don't look much different, with the accuracy still being 0.5 - effectively random.
Step 2: Detect Dogs¶
In this section, we use a pre-trained ResNet-50 model to detect dogs in images. Our first line of code downloads the ResNet-50 model, along with weights that have been trained on ImageNet, a very large, very popular dataset used for image classification and other vision tasks. ImageNet contains over 10 million URLs, each linking to an image containing an object from one of 1000 categories. Given an image, this pre-trained ResNet-50 model returns a prediction (derived from the available categories in ImageNet) for the object that is contained in the image.
from keras.applications.resnet50 import ResNet50
# define ResNet50 model
ResNet50_model = ResNet50(weights='imagenet')
Pre-process the Data¶
When using TensorFlow as backend, Keras CNNs require a 4D array (which we'll also refer to as a 4D tensor) as input, with shape
$$ (\text{nb_samples}, \text{rows}, \text{columns}, \text{channels}), $$where nb_samples corresponds to the total number of images (or samples), and rows, columns, and channels correspond to the number of rows, columns, and channels for each image, respectively.
The path_to_tensor function below takes a string-valued file path to a color image as input and returns a 4D tensor suitable for supplying to a Keras CNN. The function first loads the image and resizes it to a square image that is $224 \times 224$ pixels. Next, the image is converted to an array, which is then resized to a 4D tensor. In this case, since we are working with color images, each image has three channels. Likewise, since we are processing a single image (or sample), the returned tensor will always have shape
The paths_to_tensor function takes a numpy array of string-valued image paths as input and returns a 4D tensor with shape
Here, nb_samples is the number of samples, or number of images, in the supplied array of image paths. It is best to think of nb_samples as the number of 3D tensors (where each 3D tensor corresponds to a different image) in your dataset!
from keras.preprocessing import image
from tqdm import tqdm
def path_to_tensor(img_path):
# loads RGB image as PIL.Image.Image type
img = image.load_img(img_path, target_size=(224, 224))
# convert PIL.Image.Image type to 3D tensor with shape (224, 224, 3)
x = image.img_to_array(img)
# convert 3D tensor to 4D tensor with shape (1, 224, 224, 3) and return 4D tensor
return np.expand_dims(x, axis=0)
def paths_to_tensor(img_paths):
list_of_tensors = [path_to_tensor(img_path) for img_path in tqdm(img_paths)]
return np.vstack(list_of_tensors)
Making Predictions with ResNet-50¶
Getting the 4D tensor ready for ResNet-50, and for any other pre-trained model in Keras, requires some additional processing. First, the RGB image is converted to BGR by reordering the channels. All pre-trained models have the additional normalization step that the mean pixel (expressed in RGB as $[103.939, 116.779, 123.68]$ and calculated from all pixels in all images in ImageNet) must be subtracted from every pixel in each image. This is implemented in the imported function preprocess_input. If you're curious, you can check the code for preprocess_input here.
Now that we have a way to format our image for supplying to ResNet-50, we are now ready to use the model to extract the predictions. This is accomplished with the predict method, which returns an array whose $i$-th entry is the model's predicted probability that the image belongs to the $i$-th ImageNet category. This is implemented in the ResNet50_predict_labels function below.
By taking the argmax of the predicted probability vector, we obtain an integer corresponding to the model's predicted object class, which we can identify with an object category through the use of this dictionary.
from keras.applications.resnet50 import preprocess_input, decode_predictions
def ResNet50_predict_labels(img_path):
# returns prediction vector for image located at img_path
img = preprocess_input(path_to_tensor(img_path))
return np.argmax(ResNet50_model.predict(img))
Write a Dog Detector¶
While looking at the dictionary, you will notice that the categories corresponding to dogs appear in an uninterrupted sequence and correspond to dictionary keys 151-268, inclusive, to include all categories from 'Chihuahua' to 'Mexican hairless'. Thus, in order to check to see if an image is predicted to contain a dog by the pre-trained ResNet-50 model, we need only check if the ResNet50_predict_labels function above returns a value between 151 and 268 (inclusive).
We use these ideas to complete the dog_detector function below, which returns True if a dog is detected in an image (and False if not).
### returns "True" if a dog is detected in the image stored at img_path
def dog_detector(img_path):
prediction = ResNet50_predict_labels(img_path)
#print if any human resembles a canine
print(img_path) if ((prediction <= 268) & (prediction >= 151)) and 'lfw' in img_path else None
return ((prediction <= 268) & (prediction >= 151))
(IMPLEMENTATION) Assess the Dog Detector¶
Question 3: Use the code cell below to test the performance of your dog_detector function.
- What percentage of the images in
human_files_shorthave a detected dog? - What percentage of the images in
dog_files_shorthave a detected dog?
Answer:
- What percentage of the images in
human_files_shorthave a detected dog?Only 1% human faces were incorrectly reccognized as dog-face Different runs with different random seed turn up 4 issues... - What percentage of the images in
dog_files_shorthave a detected dog?100% of the dog faces were detected correctly by the ResNet-50 Dog Detector.
### TODO: Test the performance of the dog_detector function
### on the images in human_files_short and dog_files_short.
hres_det = list(map(dog_detector, human_files_short))
dres_det = list(map(dog_detector, dog_files_short))
print ('{}\n{}'.format(sum(hres_det),sum(dres_det)))
Step 3: Create a CNN to Classify Dog Breeds (from Scratch)¶
Now that we have functions for detecting humans and dogs in images, we need a way to predict breed from images. In this step, you will create a CNN that classifies dog breeds. You must create your CNN from scratch (so, you can't use transfer learning yet!), and you must attain a test accuracy of at least 1%. In Step 5 of this notebook, you will have the opportunity to use transfer learning to create a CNN that attains greatly improved accuracy.
Be careful with adding too many trainable layers! More parameters means longer training, which means you are more likely to need a GPU to accelerate the training process. Thankfully, Keras provides a handy estimate of the time that each epoch is likely to take; you can extrapolate this estimate to figure out how long it will take for your algorithm to train.
We mention that the task of assigning breed to dogs from images is considered exceptionally challenging. To see why, consider that even a human would have great difficulty in distinguishing between a Brittany and a Welsh Springer Spaniel.
| Brittany | Welsh Springer Spaniel |
|---|---|
 |
 |
It is not difficult to find other dog breed pairs with minimal inter-class variation (for instance, Curly-Coated Retrievers and American Water Spaniels).
| Curly-Coated Retriever | American Water Spaniel |
|---|---|
 |
 |
Likewise, recall that labradors come in yellow, chocolate, and black. Your vision-based algorithm will have to conquer this high intra-class variation to determine how to classify all of these different shades as the same breed.
| Yellow Labrador | Chocolate Labrador | Black Labrador |
|---|---|---|
 |
 |
 |
We also mention that random chance presents an exceptionally low bar: setting aside the fact that the classes are slightly imabalanced, a random guess will provide a correct answer roughly 1 in 133 times, which corresponds to an accuracy of less than 1%.
Remember that the practice is far ahead of the theory in deep learning. Experiment with many different architectures, and trust your intuition. And, of course, have fun!
Pre-process the Data¶
We rescale the images by dividing every pixel in every image by 255.
from PIL import ImageFile
ImageFile.LOAD_TRUNCATED_IMAGES = True
# pre-process the data for Keras
train_tensors = paths_to_tensor(train_files).astype('float32')/255
valid_tensors = paths_to_tensor(valid_files).astype('float32')/255
test_tensors = paths_to_tensor(test_files).astype('float32')/255
(IMPLEMENTATION) Model Architecture¶
Create a CNN to classify dog breed. At the end of your code cell block, summarize the layers of your model by executing the line:
model.summary()
We have imported some Python modules to get you started, but feel free to import as many modules as you need. If you end up getting stuck, here's a hint that specifies a model that trains relatively fast on CPU and attains >1% test accuracy in 5 epochs:
Question 4: Outline the steps you took to get to your final CNN architecture and your reasoning at each step. If you chose to use the hinted architecture above, describe why you think that CNN architecture should work well for the image classification task.
Answer:
As mentioned in the videos, the AlexNet was the first one to introduce the Conv-Pool-Dropout architecture with ReLU activation. While computationally intensive this is known to give good results in the image recognition arena. It would be instructional to try out the VGG Architecture along with this one as well.
I use the same exact architecture as suggested above, using 'valid' padding. This is different from the one that I used earlier in the Conv2D example. While keeping the layer sizes as given in the model summary above, I've added 3 Dropout layers. To begin with, I started with a value of 0.2 for the drop-rate. The results though seem to be pretty nice, giving an accuracy of 4.18%, much better than required 1%.
from keras.layers import Conv2D, MaxPooling2D, GlobalAveragePooling2D
from keras.layers import Dropout, Flatten, Dense
from keras.models import Sequential
model = Sequential()
### TODO: Define your architecture.
# Start off with the one in the other miniproject (or the same used above!)
# Then modify according to the hint given...
model = Sequential()
model.add(Conv2D(filters=16, kernel_size=2, padding='valid', activation='relu',
input_shape=(224, 224, 3)))
model.add(MaxPooling2D(pool_size=2))
model.add(Dropout(0.2))
model.add(Conv2D(filters=32, kernel_size=2, padding='valid', activation='relu'))
model.add(MaxPooling2D(pool_size=2))
model.add(Dropout(0.2))
model.add(Conv2D(filters=64, kernel_size=2, padding='valid', activation='relu'))
model.add(MaxPooling2D(pool_size=2))
model.add(Dropout(0.2))
model.add(GlobalAveragePooling2D())
model.add(Dense(133, activation='softmax'))
model.summary()
Compile the Model¶
model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])
(IMPLEMENTATION) Train the Model¶
Train your model in the code cell below. Use model checkpointing to save the model that attains the best validation loss.
You are welcome to augment the training data, but this is not a requirement.
from keras.callbacks import ModelCheckpoint
### TODO: specify the number of epochs that you would like to use to train the model.
# takes about 1:10 to run
epochs = 20
### Do NOT modify the code below this line.
checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.from_scratch.hdf5',
verbose=1, save_best_only=True)
model.fit(train_tensors, train_targets,
validation_data=(valid_tensors, valid_targets),
epochs=epochs, batch_size=20, callbacks=[checkpointer], verbose=1)
Load the Model with the Best Validation Loss¶
model.load_weights('saved_models/weights.best.from_scratch.hdf5')
Test the Model¶
Try out your model on the test dataset of dog images. Ensure that your test accuracy is greater than 1%.
# get index of predicted dog breed for each image in test set
dog_breed_predictions = [np.argmax(model.predict(np.expand_dims(tensor, axis=0))) for tensor in test_tensors]
# report test accuracy
test_accuracy = 100*np.sum(np.array(dog_breed_predictions)==np.argmax(test_targets, axis=1))/len(dog_breed_predictions)
print('Test accuracy: %.4f%%' % test_accuracy)
bottleneck_features = np.load('bottleneck_features/DogVGG16Data.npz')
train_VGG16 = bottleneck_features['train']
valid_VGG16 = bottleneck_features['valid']
test_VGG16 = bottleneck_features['test']
Model Architecture¶
The model uses the the pre-trained VGG-16 model as a fixed feature extractor, where the last convolutional output of VGG-16 is fed as input to our model. We only add a global average pooling layer and a fully connected layer, where the latter contains one node for each dog category and is equipped with a softmax.
VGG16_model = Sequential()
VGG16_model.add(GlobalAveragePooling2D(input_shape=train_VGG16.shape[1:]))
VGG16_model.add(Dense(133, activation='softmax'))
VGG16_model.summary()
Compile the Model¶
VGG16_model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy'])
Train the Model¶
from keras.callbacks import ModelCheckpoint
checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.VGG16.hdf5',
verbose=1, save_best_only=True)
VGG16_model.fit(train_VGG16, train_targets,
validation_data=(valid_VGG16, valid_targets),
epochs=20, batch_size=20, callbacks=[checkpointer], verbose=1)
Load the Model with the Best Validation Loss¶
VGG16_model.load_weights('saved_models/weights.best.VGG16.hdf5')
Test the Model¶
Now, we can use the CNN to test how well it identifies breed within our test dataset of dog images. We print the test accuracy below.
# get index of predicted dog breed for each image in test set
VGG16_predictions = [np.argmax(VGG16_model.predict(np.expand_dims(feature, axis=0))) for feature in test_VGG16]
# report test accuracy
test_accuracy = 100*np.sum(np.array(VGG16_predictions)==np.argmax(test_targets, axis=1))/len(VGG16_predictions)
print('Test accuracy: %.4f%%' % test_accuracy)
Predict Dog Breed with the Model¶
from extract_bottleneck_features import *
def VGG16_predict_breed(img_path):
# extract bottleneck features
bottleneck_feature = extract_VGG16(path_to_tensor(img_path))
# obtain predicted vector
predicted_vector = VGG16_model.predict(bottleneck_feature)
# return dog breed that is predicted by the model
return dog_names[np.argmax(predicted_vector)]
import random
randomlist = random.sample(range(len(test_files)), 5)
for idx in randomlist:
print('{:3d} {:25.25} {:25.25}'.format(idx, VGG16_predict_breed(test_files[idx]), dog_names[np.argmax(test_targets[idx])]))
Step 5: Create a CNN to Classify Dog Breeds (using Transfer Learning)¶
You will now use transfer learning to create a CNN that can identify dog breed from images. Your CNN must attain at least 60% accuracy on the test set.
In Step 4, we used transfer learning to create a CNN using VGG-16 bottleneck features. In this section, you must use the bottleneck features from a different pre-trained model. To make things easier for you, we have pre-computed the features for all of the networks that are currently available in Keras:
- VGG-19 bottleneck features
- ResNet-50 bottleneck features
- Inception bottleneck features
- Xception bottleneck features
The files are encoded as such:
Dog{network}Data.npz
where {network}, in the above filename, can be one of VGG19, Resnet50, InceptionV3, or Xception. Pick one of the above architectures, download the corresponding bottleneck features, and store the downloaded file in the bottleneck_features/ folder in the repository.
(IMPLEMENTATION) Obtain Bottleneck Features¶
In the code block below, extract the bottleneck features corresponding to the train, test, and validation sets by running the following:
bottleneck_features = np.load('bottleneck_features/Dog{network}Data.npz')
train_{network} = bottleneck_features['train']
valid_{network} = bottleneck_features['valid']
test_{network} = bottleneck_features['test']### TODO: Obtain bottleneck features from another pre-trained CNN.
# Reason to use Resnet50 - its already downloaded, Inc=1.6GB, eXc=3.1GB not done yet.
bottleneck_features = np.load('bottleneck_features/DogResnet50Data.npz')
train_Resnet50 = bottleneck_features['train']
valid_Resnet50 = bottleneck_features['valid']
test_Resnet50 = bottleneck_features['test']
(IMPLEMENTATION) Model Architecture¶
Create a CNN to classify dog breed. At the end of your code cell block, summarize the layers of your model by executing the line:
<your model's name>.summary()
Question 5: Outline the steps you took to get to your final CNN architecture and your reasoning at each step. Describe why you think the architecture is suitable for the current problem.
Answer:
We use transfer learning, i.e. pre-tuned set of parameters, to identify the features.
Since we used the similar architecture in the previous step, we'll re-use the same approach (and code!) with the Resnet50 dataset. In the Resnet50 model, the dog features are squeezed into a 2048 vector (GlobalAveragePooling2D) and then a fully-connected layer is used to obtain the predicted probabilities (133 possible breeds).
### TODO: Define your architecture.
Resnet50_model = Sequential()
Resnet50_model.add(GlobalAveragePooling2D(input_shape=train_Resnet50.shape[1:]))
Resnet50_model.add(Dense(133, activation='softmax'))
Resnet50_model.summary()
(IMPLEMENTATION) Compile the Model¶
### TODO: Compile the model.
Resnet50_model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy'])
(IMPLEMENTATION) Train the Model¶
Train your model in the code cell below. Use model checkpointing to save the model that attains the best validation loss.
You are welcome to augment the training data, but this is not a requirement.
### TODO: Train the model.
from keras.callbacks import ModelCheckpoint
checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.Resnet50.hdf5',
verbose=1, save_best_only=True)
Resnet50_model.fit(train_Resnet50, train_targets,
validation_data=(valid_Resnet50, valid_targets),
epochs=25, batch_size=32, callbacks=[checkpointer], verbose=1)
(IMPLEMENTATION) Load the Model with the Best Validation Loss¶
### TODO: Load the model weights with the best validation loss.
Resnet50_model.load_weights('saved_models/weights.best.Resnet50.hdf5')
(IMPLEMENTATION) Test the Model¶
Try out your model on the test dataset of dog images. Ensure that your test accuracy is greater than 60%.
### TODO: Calculate classification accuracy on the test dataset.
# get index of predicted dog breed for each image in test set
Resnet50_predictions = [np.argmax(Resnet50_model.predict(np.expand_dims(feature, axis=0))) for feature in test_Resnet50]
# report test accuracy
test_accuracy = 100*np.sum(np.array(Resnet50_predictions)==np.argmax(test_targets, axis=1))/len(Resnet50_predictions)
print('Test accuracy: %.4f%%' % test_accuracy)
(IMPLEMENTATION) Predict Dog Breed with the Model¶
Write a function that takes an image path as input and returns the dog breed (Affenpinscher, Afghan_hound, etc) that is predicted by your model.
Similar to the analogous function in Step 5, your function should have three steps:
- Extract the bottleneck features corresponding to the chosen CNN model.
- Supply the bottleneck features as input to the model to return the predicted vector. Note that the argmax of this prediction vector gives the index of the predicted dog breed.
- Use the
dog_namesarray defined in Step 0 of this notebook to return the corresponding breed.
The functions to extract the bottleneck features can be found in extract_bottleneck_features.py, and they have been imported in an earlier code cell. To obtain the bottleneck features corresponding to your chosen CNN architecture, you need to use the function
extract_{network}
where {network}, in the above filename, should be one of VGG19, Resnet50, InceptionV3, or Xception.
### TODO: Write a function that takes a path to an image as input
### and returns the dog breed that is predicted by the model.
from extract_bottleneck_features import *
def Resnet50_predict_breed(img_path):
# extract bottleneck features
bottleneck_feature = extract_Resnet50(path_to_tensor(img_path))
# obtain predicted vector
predicted_vector = Resnet50_model.predict(bottleneck_feature)
# return dog breed that is predicted by the model
return dog_names[np.argmax(predicted_vector)]
import random
randomlist = random.sample(range(len(test_files)), 5)
for index, idx in enumerate(randomlist, start=1):
print('{} > {:3d} {:25.25} {:25.25}'.format(index, idx, Resnet50_predict_breed(test_files[idx]), dog_names[np.argmax(test_targets[idx])]))
Step 6: Write your Algorithm¶
Write an algorithm that accepts a file path to an image and first determines whether the image contains a human, dog, or neither. Then,
- if a dog is detected in the image, return the predicted breed.
- if a human is detected in the image, return the resembling dog breed.
- if neither is detected in the image, provide output that indicates an error.
You are welcome to write your own functions for detecting humans and dogs in images, but feel free to use the face_detector and dog_detector functions developed above. You are required to use your CNN from Step 5 to predict dog breed.
Some sample output for our algorithm is provided below, but feel free to design your own user experience!
(IMPLEMENTATION) Write your Algorithm¶
### TODO: Write your algorithm.
### Feel free to use as many code cells as needed.
def show_img(imagefile):
img = cv2.imread(imagefile)
cv_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
plt.imshow(cv_rgb)
plt.show()
def get_dog_breed(imagefile):
# Same as Resnet50_predict_breed, but dont wrap
# extract bottleneck features
bottleneck_feature = extract_Resnet50(path_to_tensor(imagefile))
# obtain predicted vector
predicted_vector = Resnet50_model.predict(bottleneck_feature)
sorted_vector = predicted_vector.argsort()[::-1]
sorted_probs = predicted_vector[0,sorted_vector]
sorted_names = np.array(dog_names)[sorted_vector]
return sorted_names[:, ::-1][:, :3], sorted_probs[:, ::-1][:, :3]
def check_if_human(imagefile):
num_faces = face_detector(imagefile)
return num_faces > 0
def check_if_dog(imagefile):
is_dog = dog_detector(imagefile)
return is_dog
print('done')
human_dog_misclass=['lfw/Andrew_Fastow/Andrew_Fastow_0001.jpg',\
'lfw/George_W_Bush/George_W_Bush_0187.jpg',\
'lfw/Doris_Schroeder/Doris_Schroeder_0001.jpg',\
'lfw/Roy_Williams/Roy_Williams_0001.jpg']
for idx, imgfile in enumerate(human_files_short):
print(idx, imgfile) if imgfile in human_dog_misclass else None
print('done')
print('\n============\n')
show_img(human_files_short[75])
human = check_if_human(human_files_short[75])
print('Human? = {}'.format(human))
dog = check_if_dog(human_files_short[75])
print('Dog? = {}'.format(dog))
pred_name, pred_prob = get_dog_breed(human_files_short[75])
print(pred_name[0])
print(['%5.3f' % val for val in pred_prob[0]])
print('\n============\n')
show_img(human_files_short[17])
human = check_if_human(human_files_short[17])
print('Human? = {}'.format(human))
dog = check_if_dog(human_files_short[17])
print('Dog? = {}'.format(dog))
pred_name, pred_prob = get_dog_breed(human_files_short[17])
print(pred_name)
print(['%5.3f' % val for val in pred_prob[0]])
print('\n============\n')
show_img(human_files_short[39])
human = check_if_human(human_files_short[39])
print('Human? = {}'.format(human))
dog = check_if_dog(human_files_short[39])
print('Dog? = {}'.format(dog))
pred_name, pred_prob = get_dog_breed(human_files_short[39])
print(pred_name)
print(['%5.3f' % val for val in pred_prob[0]])
print('\n============\n')
show_img(human_files_short[19])
human = check_if_human(human_files_short[19])
print('Human? = {}'.format(human))
dog = check_if_dog(human_files_short[19])
print('Dog? = {}'.format(dog))
pred_name, pred_prob = get_dog_breed(human_files_short[19])
print(pred_name)
print(['%5.3f' % val for val in pred_prob[0]])
print('\n============\n')
show_img(dog_files_short[1])
human = check_if_human(dog_files_short[1])
print('Human? = {}'.format(human))
dog = check_if_dog(dog_files_short[1])
print('Dog? = {}'.format(dog))
pred_name, pred_prob = get_dog_breed(dog_files_short[1])
print(pred_name)
print(['%5.3f' % val for val in pred_prob[0]])
print('\n============\n')
show_img(dog_files_short[19])
human = check_if_human(dog_files_short[19])
print('Human? = {}'.format(human))
dog = check_if_dog(dog_files_short[19])
print('Dog? = {}'.format(dog))
pred_name, pred_prob = get_dog_breed(dog_files_short[19])
print(pred_name)
print(['%5.3f' % val for val in pred_prob[0]])
Step 7: Test Your Algorithm¶
In this section, you will take your new algorithm for a spin! What kind of dog does the algorithm think that you look like? If you have a dog, does it predict your dog's breed accurately? If you have a cat, does it mistakenly think that your cat is a dog?
(IMPLEMENTATION) Test Your Algorithm on Sample Images!¶
Test your algorithm at least six images on your computer. Feel free to use any images you like. Use at least two human and two dog images.
Question 6: Is the output better than you expected :) ? Or worse :( ? Provide at least three possible points of improvement for your algorithm.
Answer:
The output is within the limits of error expected. Some humans were detected as dogs, and some dogs are misidentified as humans. However, non-human and non-dog images were correcctly identified as not containing either.
Suggested improvements:
Expand the last layer to 133+1 (for human) or 133+2 ( 1 for human, the other for non-dog-non-human) and re-train the network with the corpus of dog and human images available.
Additionally, (and this is probably a terrrible idea) we can expand the last layer to 133+ 5748 (number of individuals images) so it can identify the humans with their names as well.We havn't used image augmentation here like in the optional part. Using that, on both human and dog images and re-traininnng can improve the accuracy.
This approach will probably negate the benifits of transfer learning..? We can train a new classifier of the type (A + B) where A = trained weights of dog images set, and B = trained weights of human images set.Algorithm tuning: For example, weight and bias initialization from a truncated gaussian distribution have been proved to be essential to the training performance in several other projects. A re-look at initializing the weights based on the following would possibly help.
https://machinelearningmastery.com/improve-deep-learning-performance/ http://deepdish.io/2015/02/24/network-initialization/
## TODO: Execute your algorithm from Step 6 on
## at least 6 images on your computer.
## Feel free to use as many code cells as needed.
def make_pred_str(nm, pr):
pstr = 'Dog detected. Most likely breed: '
for idx in range(1):
pstr += str(nm[0][idx]) + ': ' + '{:.5f}'.format(pr[0][idx]) + '\t'
return pstr +'\n'
def whosit(somepic):
pred_str = ''
print(somepic)
show_img(somepic)
human = check_if_human(somepic)
print('Is human? = {}'.format(human))
dog = check_if_dog(somepic)
print('Is dog? = {}'.format(dog))
if dog:
pred_name, pred_prob = get_dog_breed(somepic)
pred_str = make_pred_str(pred_name, pred_prob)
print(pred_str)
if human:
print('Human detected.')
if not (dog or human):
print("Could not detect either human or dog...")
my_files = np.array(glob("testimages/*"))
my_files = np.sort(my_files)
#print(my_files)
for apic in my_files:
whosit(apic)
print(apic)
print('============\n\n')
#print(pred_str(pred_name, pred_prob))
